Global Model for Failure Prediction for Artificial Lift Systems

1. A method for predicting failures in an artificial lift system, the method comprising: extracting one or more features from a dataset including time sampled performance of a plurality of artificial lift systems disposed across a plurality of different oil fields, the dataset including data from failed and normally operating artificial lift systems; identifying pre-failure signatures based at least in part on a moving window of operational data in the extracted features preceding a known failure; forming a learning model based on identified pre-failure signatures in the extracted features, the learning model configured to predict a failure of an artificial lift system based on observation of one of the identified pre-failure signatures in operational data received from the artificial lift system; and predicting one or more failures in an artificial lift system based on the learning model.

2. The method of claim 1 , wherein forming the learning model includes labeling the extracted features to define interrelationships among the features included in the dataset.

3. The method of claim 2 , wherein forming the learning model includes training a multi-class support vector machine using the labeled extracted features.

4. The method of claim 1 , wherein extracting one or more features includes applying a moving median feature extraction process to the dataset.

5. The method of claim 4 , wherein the moving median feature extraction process calculates a global median, a mid-term performance median, and a current performance median.

6. The method of claim 5 , wherein extracting the one or more features includes calculating features by dividing at least one of the mid-term performance median or the current performance median by the global median.

7. The method of claim 1 , wherein the dataset includes measurements of attributes from a plurality of pump off controllers.

8. The method of claim 7 , wherein the attributes are selected from a group of attributes consisting of: card area; peak surface load; minimum surface load; strokes per minute; surface stroke length; flow line pressure; pump fillage; prior day cycles; and daily run time.

9. The method of claim 8 , further comprising calculating a card unchanged days attribute based on a number of days of one or more unchanged attributes received from a pump off controller.

10. The method of claim 8 , further comprising calculating a daily runtime ratio based on the daily run time.

11. The method of claim 1 , further comprising evaluating a precision of the failure prediction.

12. The method of claim 1 , further comprising periodically updating the learning model with a refreshed dataset including time sampled performance of the plurality of artificial lift systems.

13. A computer-readable medium having computer-executable instructions stored thereon which, when executed by a computing system, cause the computing system to perform a method for predicting failures in an artificial lift system, the method comprising: extracting one or more features from a dataset including time sampled performance of a plurality of artificial lift systems disposed across a plurality of different oil fields, the dataset including data from failed and normally operating artificial lift systems; identifying pre-failure signatures based at least in part on a moving window of operational data in the extracted features preceding a known failure; forming a learning model based on identified pre-failure signatures in the extracted features, the learning model configured to predict a failure of an artificial lift system based on observation of one of the identified pre-failure signatures in operational data received from the artificial lift system; and predicting one or more failures in an artificial lift system based on the learning model.

14. A system for predicting failures in an artificial lift system, the system comprising: a processor; a memory communicatively connected to the processor and storing computer-executable instructions that, when executed by the processor, cause the system to: receive a dataset of time-sampled data from each of a plurality of artificial lift systems disposed across a plurality of different oil fields, the dataset including data from failed and normally operating artificial lift systems; receive data labels from a user, the data labels defining one or more types of failures of artificial lift systems; identify pre-failure signatures based at least in part on a moving window of operational data in the extracted features preceding a known failure included in the one or more types of failures; generate a learning model by a multi-class support vector machine based on the labeled data, the learning model including one or more identified pre-failure signatures and predict a failure of an artificial lift system based on observation of one of the identified pre-failure signatures in operational data received from the artificial lift system.

15. The system of claim 14 , wherein the plurality of artificial lift systems disposed across a plurality of different oil fields include one or more artificial lift systems selected from a group of systems consisting of: a gas lift; a hydraulic pumping unit; an electric submersible pump; a progressive cavity pump; and a rod pump.

16. The system of claim 14 , wherein the system is further configured to extract one or more features from a dataset based on labels applied to the time-sampled data.

17. The system of claim 14 , wherein updated data is provided to the multi-class support vector machine to generate an updated learning model.

18. The system of claim 14 , wherein the system is further configured to apply a clustering algorithm comprising: collecting data representing artificial lift system failures having a common failure type into a first cluster; and determining one or more pre-failure signatures in the time-sampled data.

19. The system of claim 18 , wherein the clustering algorithm further includes collecting data representing normal operation of an artificial lift system into a second cluster.

20. The system of claim 14 , wherein the dataset includes measurements of attributes, and wherein the attributes are selected from a group of attributes consisting of: card area; peak surface load; minimum surface load; strokes per minute; surface stroke length; flow line pressure; pump fillage; prior day cycles; and daily run time.

21. The method of claim 1 , wherein the plurality of artificial lift systems disposed across a plurality of different oil fields include one or more artificial lift systems selected from a group of systems consisting of: a gas lift; a hydraulic pumping unit; an electric submersible pump; a progressive cavity pump; and a rod pump.